Akka（36）： Http：Client-side-Api，Client-Connections

Posted 2020-10-13 雪川大虫

   Akka-http的客户端Api应该是以HttpRequest操作为主轴的网上消息交换模式编程工具。我们知道：Akka-http是搭建在Akka-stream之上的。所以，Akka-http在客户端构建与服务器的连接通道也可以用Akka-stream的Flow来表示。这个Flow可以通过调用Http.outgoingConnection来获取：

  /**
   * Creates a [[akka.stream.scaladsl.Flow]] representing a prospective HTTP client connection to the given endpoint.
   * Every materialization of the produced flow will attempt to establish a new outgoing connection.
   *
   * To configure additional settings for requests made using this method,
   * use the `akka.http.client` config section or pass in a [[akka.http.scaladsl.settings.ClientConnectionSettings]] explicitly.
   */
  def outgoingConnection(host: String, port: Int = 80,
                         localAddress: Option[InetSocketAddress] = None,
                         settings:     ClientConnectionSettings  = ClientConnectionSettings(system),
                         log:          LoggingAdapter            = system.log): Flow[HttpRequest, HttpResponse, Future[OutgoingConnection]] =
    _outgoingConnection(host, port, settings.withLocalAddressOverride(localAddress), ConnectionContext.noEncryption(), ClientTransport.TCP, log)

我们看到：这个函数实现了对Server端地址host+port的设定，返回的结果类型是Flow[HttpRequest,HttpResponse,Future[OutgoingConnection]]。这个Flow代表将输入的HttpRequest转换成输出的HttpResponse。这个转换过程包括了与Server之间的Request,Response消息交换。下面我们试着用这个Flow来向Server端发送request，并获取response：

  val connFlow: Flow[HttpRequest,HttpResponse,Future[Http.OutgoingConnection]] =
    Http().outgoingConnection("akka.io")
  
  def sendHttpRequest(req: HttpRequest) = {
    Source.single(req)
      .via(connFlow)
      .runWith(Sink.head)
  }
  
  sendHttpRequest(HttpRequest(uri="/"))
      .andThen{
        case Success(resp) => println(s"got response: ${resp.status.intValue()}")
        case Failure(err) => println(s"request failed: ${err.getMessage}")
      }
     .andThen {case _ => sys.terminate()}

上面的这种模式就是所谓Connection-Level-Client-Side-Api。这种模式可以让用户有更大程度的自由度控制connection的构建、使用及在connection上发送request的方式。一般来讲，当返回response的entity被完全消耗后系统会自动close connection，这套api还提供了一些手动方法可以在有需要的情况下手动进行connection close，如下：

 //close connection by cancelling response entity
  resp.entity.dataBytes.runWith(Sink.cancelled)
 //close connection by receiving response with close header
  Http().bindAndHandleSync(
    { req ? HttpResponse(headers = headers.Connection("close") :: Nil) },
    "akka.io",
    80)(mat)

Akka-http客户端api还有一种实用的Host-Level-Client-Side-Api模式。这套api能自动针对每个端点维护一个连接池（connection-pool），用户只需对连接池进行配置。系统按照连接池配置自动维护池内线程的生、死、动、停。akka-http.host-connection-pool配置中max-connections,max-open-requests,pipelining-limit等控制着connection、在途request的数量，需要特别注意。针对某个端点的连接池是通过Http().cachedHostConnectionPool(endPoint)获取的。同样，获取的也是一个client-flow实例。因为系统自动维护着线程池，所以client-flow实例可以任意引用，无论调用次数与调用时间间隔。cachedHostConnectionPool()函数定义如下：

  /**
   * Same as [[#cachedHostConnectionPool]] but for encrypted (HTTPS) connections.
   *
   * If an explicit [[ConnectionContext]] is given then it rather than the configured default [[ConnectionContext]] will be used
   * for encryption on the connections.
   *
   * To configure additional settings for the pool (and requests made using it),
   * use the `akka.http.host-connection-pool` config section or pass in a [[ConnectionPoolSettings]] explicitly.
   */
  def cachedHostConnectionPoolHttps[T](host: String, port: Int = 443,
                                       connectionContext: HttpsConnectionContext = defaultClientHttpsContext,
                                       settings:          ConnectionPoolSettings = defaultConnectionPoolSettings,
                                       log:               LoggingAdapter         = system.log)(implicit fm: Materializer): Flow[(HttpRequest, T), (Try[HttpResponse], T), HostConnectionPool] = {
    val cps = ConnectionPoolSetup(settings, connectionContext, log)
    val setup = HostConnectionPoolSetup(host, port, cps)
    cachedHostConnectionPool(setup)
  }

函数返回结果类型：Flow[(HttpRequest,T),(Try[HttpResponse],T),HostConnectionPool]。因为线程池内的线程是异步构建request和接收response的，而返回response的顺序未必按照发送request的顺序，所以需要一个tuple2的T类型标示request与返回的response进行匹配。线程池会根据idle-timeout自动终止，也可以手动通过HostConnectionPool.shutDown()实现：

  /**
   * Represents a connection pool to a specific target host and pool configuration.
   */
  final case class HostConnectionPool private[http] (setup: HostConnectionPoolSetup)(
    private[http] val gateway: PoolGateway) { // enable test access

    /**
     * Asynchronously triggers the shutdown of the host connection pool.
     *
     * The produced [[scala.concurrent.Future]] is fulfilled when the shutdown has been completed.
     */
    def shutdown()(implicit ec: ExecutionContextExecutor): Future[Done] = gateway.shutdown()

    private[http] def toJava = new akka.http.javadsl.HostConnectionPool {
      override def setup = HostConnectionPool.this.setup
      override def shutdown(executor: ExecutionContextExecutor): CompletionStage[Done] = HostConnectionPool.this.shutdown()(executor).toJava
    }
  }

也可以通过Http().shutdownAllConnectionPools()一次性终止ActorSystem内所有线程池:

  /**
   * Triggers an orderly shutdown of all host connections pools currently maintained by the [[akka.actor.ActorSystem]].
   * The returned future is completed when all pools that were live at the time of this method call
   * have completed their shutdown process.
   *
   * If existing pool client flows are re-used or new ones materialized concurrently with or after this
   * method call the respective connection pools will be restarted and not contribute to the returned future.
   */
  def shutdownAllConnectionPools(): Future[Unit] = {
    val shutdownCompletedPromise = Promise[Done]()
    poolMasterActorRef ! ShutdownAll(shutdownCompletedPromise)
    shutdownCompletedPromise.future.map(_ ? ())(system.dispatcher)
  }

我们用cachedHostConnectionPool获取一个client-flow实例：

Flow[(HttpRequest,T),(Try[HttpResponse],T),HostConnectionPool]后就可以进行输入HttpRequest到HttpResponse的转换处理。如下面的例子：

  val pooledFlow: Flow[(HttpRequest,Int),(Try[HttpResponse],Int),Http.HostConnectionPool] =
      Http().cachedHostConnectionPool[Int](host="akka.io",port=80)

  def sendPoolRequest(req: HttpRequest, marker: Int) = {
    Source.single(req -> marker)
      .via(pooledFlow)
      .runWith(Sink.head)
  }

  sendPoolRequest(HttpRequest(uri="/"), 1)
    .andThen{
      case Success((tryResp, mk)) =>
        tryResp match {
          case Success(resp) => println(s"got response: ${resp.status.intValue()}")
          case Failure(err) => println(s"request failed: ${err.getMessage}")
        }
      case Failure(err) => println(s"request failed: ${err.getMessage}")
    }
    .andThen {case _ => sys.terminate()}

在以上这个例子里实际同样会遇到Connection-Level-Api所遇的的问题，这是因为获取的线程池内的线程还是有限的，只能缓解因为request速率超出response速率所造成的request积压。目前最有效的方法还是通过使用一个queue来暂存request后再逐个处理：

    val QueueSize = 10
    // This idea came initially from this blog post:
    // http://kazuhiro.github.io/scala/akka/akka-http/akka-streams/2016/01/31/connection-pooling-with-akka-http-and-source-queue.html
    val poolClientFlow = Http().cachedHostConnectionPool[Promise[HttpResponse]]("akka.io")
    val queue =
      Source.queue[(HttpRequest, Promise[HttpResponse])](QueueSize, OverflowStrategy.dropNew)
        .via(poolClientFlow)
        .toMat(Sink.foreach({
          case ((Success(resp), p)) => p.success(resp)
          case ((Failure(e), p))    => p.failure(e)
        }))(Keep.left)
        .run()

    def queueRequest(request: HttpRequest): Future[HttpResponse] = {
      val responsePromise = Promise[HttpResponse]()
      queue.offer(request -> responsePromise).flatMap {
        case QueueOfferResult.Enqueued    => responsePromise.future
        case QueueOfferResult.Dropped     => Future.failed(new RuntimeException("Queue overflowed. Try again later."))
        case QueueOfferResult.Failure(ex) => Future.failed(ex)
        case QueueOfferResult.QueueClosed => Future.failed(new RuntimeException("Queue was closed (pool shut down) while running the request. Try again later."))
      }
    }

    val responseFuture: Future[HttpResponse] = queueRequest(HttpRequest(uri = "/"))
    responseFuture.andThen {
      case Success(resp) => println(s"got response: ${resp.status.intValue()}")
      case Failure(err) => println(s"request failed: ${err.getMessage}")
    }.andThen {case _ => sys.terminate()}

下面是本次Akka-http-client-side-connection讨论的示范源代码：

import akka.actor._
import akka.http.javadsl.{HostConnectionPool, OutgoingConnection}
import akka.stream._
import akka.stream.scaladsl._
import akka.http.scaladsl.Http
import akka.http.scaladsl.model._

import scala.concurrent._
import scala.util._

object ClientApiDemo extends App {
  implicit val sys = ActorSystem("ClientSys")
  implicit val mat = ActorMaterializer()
  implicit val ec = sys.dispatcher
/*
  val connFlow: Flow[HttpRequest,HttpResponse,Future[Http.OutgoingConnection]] =
    Http().outgoingConnection("www.sina.com")

  def sendHttpRequest(req: HttpRequest) = {
    Source.single(req)
      .via(connFlow)
      .runWith(Sink.head)
  }

  sendHttpRequest(HttpRequest(uri="/"))
      .andThen{
        case Success(resp) =>
          //close connection by cancelling response entity
          resp.entity.dataBytes.runWith(Sink.cancelled)
          println(s"got response: ${resp.status.intValue()}")
        case Failure(err) => println(s"request failed: ${err.getMessage}")
      }
  //   .andThen {case _ => sys.terminate()}


 //close connection by receiving response with close header
  Http().bindAndHandleSync(
    { req ? HttpResponse(headers = headers.Connection("close") :: Nil) },
    "akka.io",
    80)(mat)

  val pooledFlow: Flow[(HttpRequest,Int),(Try[HttpResponse],Int),Http.HostConnectionPool] =
      Http().cachedHostConnectionPool[Int](host="akka.io",port=80)

  def sendPoolRequest(req: HttpRequest, marker: Int) = {
    Source.single(req -> marker)
      .via(pooledFlow)
      .runWith(Sink.head)
  }

  sendPoolRequest(HttpRequest(uri="/"), 1)
    .andThen{
      case Success((tryResp, mk)) =>
        tryResp match {
          case Success(resp) => println(s"got response: ${resp.status.intValue()}")
          case Failure(err) => println(s"request failed: ${err.getMessage}")
        }
      case Failure(err) => println(s"request failed: ${err.getMessage}")
    }
    .andThen {case _ => sys.terminate()}
*/

    val QueueSize = 10
    // This idea came initially from this blog post:
    // http://kazuhiro.github.io/scala/akka/akka-http/akka-streams/2016/01/31/connection-pooling-with-akka-http-and-source-queue.html
    val poolClientFlow = Http().cachedHostConnectionPool[Promise[HttpResponse]]("akka.io")
    val queue =
      Source.queue[(HttpRequest, Promise[HttpResponse])](QueueSize, OverflowStrategy.dropNew)
        .via(poolClientFlow)
        .toMat(Sink.foreach({
          case ((Success(resp), p)) => p.success(resp)
          case ((Failure(e), p))    => p.failure(e)
        }))(Keep.left)
        .run()

    def queueRequest(request: HttpRequest): Future[HttpResponse] = {
      val responsePromise = Promise[HttpResponse]()
      queue.offer(request -> responsePromise).flatMap {
        case QueueOfferResult.Enqueued    => responsePromise.future
        case QueueOfferResult.Dropped     => Future.failed(new RuntimeException("Queue overflowed. Try again later."))
        case QueueOfferResult.Failure(ex) => Future.failed(ex)
        case QueueOfferResult.QueueClosed => Future.failed(new RuntimeException("Queue was closed (pool shut down) while running the request. Try again later."))
      }
    }

    val responseFuture: Future[HttpResponse] = queueRequest(HttpRequest(uri = "/"))
    responseFuture.andThen {
      case Success(resp) => println(s"got response: ${resp.status.intValue()}")
      case Failure(err) => println(s"request failed: ${err.getMessage}")
    }.andThen {case _ => sys.terminate()}

}